Wind being one of the most utilized renewable energy sources, the researchers aimed towards development of new techniques to harness it efficiently. The complex nature of wind characteristics presents hurdle for its accurate prediction. As wind speed has a close link with the power production capability of a wind farm, accurate prediction of it is the need-of-the-hour for realistic studies. In this work, long short-term memory networks (LSTMs) have been utilized by the authors for accurately forecasting wind behaviour characterized by its speed and direction. The proposed work brings novelty in the evolutionary approach for the automated design of LSTMs, which enables the estimation of the hyper-parameters in them, including their architecture, activation function, and backpropagation length, without the intervention of heuristics. The optimally designed LSTMs are trained using real wind data with very high accuracy. The forecasted data with optimal LSTMs are validated with the available original data, which shows the credibility of the proposed methodology.